Instrument Science Report ACS 2020-02

New and Improved Saturated PixelFlagging for the ACS/WFC

Yotam CohenNorman A. Grogin

January 6, 2020

Abstract

Accurate characterization of the saturation level of the ACS/WFC CCD is crucial forproper flagging of affected pixels, which users and calibration routines require knowledge of.In this work, we present a new analysis of the saturation level that offers significantimprovements and advantages over previously used methods. Unlike previous work, wemeasure the onset of saturation directly by identifying the precise charge level at which thebrightest pixel of point sources begins to spill charge into neighboring pixels. This results ina sharp decrease in the fraction of charge contained in the central pixel, coupled with asharp increase in the fraction of charge contained in neighboring pixels. Through thisanalysis, we find that the saturation level has a strong spatial dependence over the detectorarea and exhibits significant (± ∼ 6% about the mean) variations, in agreement withprevious work. Despite this qualitative agreement, we find that the saturation level currentlyused in the CALACS calibration pipeline to flag affected pixels is much too high, causing itto routinely miss many clearly saturated pixels. When using our new saturation map toperform the flagging, we find visually superior results and as many as ∼ 15% more pixelsbeing flagged as saturated in any given frame. We announce plans to implement our newsaturation map into CALACS, and discuss extensions of this work.

Copyright c©2020 The Association of Universities for Research in Astronomy, Inc. All Rights Reserved.

1 Introduction

Accurate characterization of the saturation level of the ACS/WFC CCD is crucial forproper flagging of affected pixels, which users and calibration routines require knowledge of.In this context, saturation can be defined as the charge level (in electrons) above which aparticular pixel loses its ability to linearly integrate more charge, and the pixel starts to spillsome of the excess charge into its neighboring pixels (usually called ‘bleeding’ or ‘blooming’),discussed further in Section 2. This renders both the saturated pixels and all neighboringpixels suspect for calibration and science calculations at best, and completely unusable atworst.

Early ACS work on this topic was carried out by Gilliland (2004), who used charac-teristics of saturated star images to estimate the saturation level, though only indirectly.Following SM4, a new measurement of the saturation level was performed using photontransfer curve analysis (Golimowski et al., 2011), but the measured values were only accu-rate to within ± ∼ 15% due to relatively sparse sampling. Uncertainties that large in thismeasurement has evidently led to misbehaved saturation flagging (e.g. middle right columnof Figure 5). Also, despite both previous analyses finding a position-dependent saturationlevel with significant variations (±6% about the mean) over the detector area, only a singlescalar threshold has ever been used for saturation flagging in the ACS/WFC calibrationpipeline. In this work, we present a new analysis of the saturation level which offers im-provements and advantages over both previous methods. We show that using our newlycharacterized saturation level, including its detector position dependence, provides superiorsaturated pixel flagging compared to the current uniform threshold. We discuss plans toimplement the saturation map and flagging routine into the calibration pipeline (hereafter“CALACS”).

Note that there are two types of saturation discussed in the HST (and general CCD)literature and documentation (Ryon et al., 2019): analog-to-digital (A-to-D) converter satu-ration, and full-well saturation. In this work we are concerned only with full-well saturation,which can only occur for gain settings (header keyword CCDGAIN) of 1.4 and higher for theWFC CCD. However, since CCDGAIN=2 is the only currently supported setting, and theone used by the large majority of archival data, we restrict our analysis and results to imagesusing CCDGAIN=2. We briefly discuss the impact on other gain settings in Section 3.2.

2 An Understanding and Definition of Saturation

We first examine a set of WFC pixels that we dub ‘super-hot’ pixels. These are a subsetof ordinary hot pixels (Riess et al., 2002; Ogaz et al., 2015), with the added definition thattheir dark current is sufficiently high to cause them to be saturated in the typical WFClong darks (∼ 1000 s); some have dark currents as high as ∼ 2000 electrons per second,as measured from the WFC short darks (∼ 8 s total dark time). Cutouts of a small regionsurrounding four of these super-hot pixels as seen in recent short and long super-darks (Lucaset al., 2018) are shown in Figure 1. At least a dozen of these super-hot pixels appear tobe permanent (likely manufacturing defects) and are present in all WFC frames since ACSinstallation. In addition to those, as the detector has aged, there have been an increasing

2

Figure 1: Cutouts (21×41 pixels) of recent short (top row and colorbar) and long (bottom row and colorbar)superdarks centered on four ‘super-hot’ pixels (defined in the text), each located on a different quadrant ofthe full WFC frame. The source super-hot pixel (the location of which is first identified in the short darks)is indicated with a red cross, and the serial and parallel readout directions are indicated with the red arrows.Charge seen in pixels other than the source pixel result from various different effects: familiar vertical chargeblooming in the column of the source pixel; CTE loss trails in both the serial and parallel directions oppositeto readout; and a newly documented effect (presumably horizontal charge blooming) resulting in additionalcharge leaking into either or both columns flanking the central hot column.

number of temporary or newly-permanent super-hot pixels appearing in any given annealcycle, with as many as ∼ 125 identified in some of the most recent (late 2019) darks. Thesepixels and their bleed trails are obviously a nuisance for science users of the instrument (e.g.Miles (2018)). However, they provide an extremely unique and useful opportunity to studysaturation in-orbit, because: 1) all the charge originates from a single pixel, which allowsus to study the properties of bled charge into neighboring pixels decoupled from chargesfrom other sources; 2) they are not convolved with the PSF since the source of their chargeoriginates in the detector and does not pass through optical elements; 3) the dark currentis typically stable over long timescales; and 4) they appear in all dark and science frames,at identical pixel coordinates, meaning we can identify and study them with ease in tens ofthousands of images across orders of magnitude in dark time and throughout the detectorlifetime.

Making use of the last of those facts, we have plotted in Figure 2 the intensity of arepresentative super-hot pixel (blue scatter points), and the two immediately neighboringpixels in its column (red and green scatter points), as a function of dark time (integration

3

Figure 2: Intensity of a particular ‘super-hot’ pixel (blue scatter points), and its two nearest neighboringpixels in its column (red and green scatter points), as a function of dark time. The plotted scatter data aretaken from a sample of over 40,000 individual WFC exposures spanning more than three orders of magnitudein dark time. The black solid line is a piecewise fit consisting of three linear functions fit to the central super-hot pixel data in log-log space; the three pieces are constrained to be continuous at the two breakpoints.The grey, vertical, dashed lines are plotted at the abscissa of the two fitted breakpoints, and are meant todelineate the three regimes of pixel behavior which we describe in the main text.

time of dark current, which is a sum of exposure time and additional time during which darkcurrent may accumulate). The plotted measurements come from over 40,000 post-SM4 WFC(bias-subtracted) images, spanning integration times between ∼ 0.5 s to ∼ 3,000 s. The aimof this figure is to illustrate the different regimes of charge accumulation. (The behavior ofany of the other super-hot pixels and its neighbors is qualitatively the same.) The black solidline is a piecewise fit consisting of three linear functions fit to the super-hot pixel data inlog-log space; the three pieces are constrained to be continuous at the two breakpoints. Thegrey, vertical, dashed lines are plotted at the abscissa of the two fitted breakpoints, and aremeant to delineate the three regimes of charge-accumulation which we have identified: In thefirst regime, left of the first dashed line, the super-hot pixel integrates linearly (which appearscurved in the plotted semi-log space), while its two neighboring pixels are consistent with thebackground level, as expected. Immediately right of the first dashed line, the super-hot pixelbegins to bleed some, but not all, further accumulated charge into its neighboring pixels,causing the sudden increase in their intensity, and the sub-linear integration of the super-hot pixel; this behavior continues until the second dashed line. Finally, right of the seconddashed line, the integration curve of the source pixel shallows once again to a vanishinglysmall slope, asymptotically approaching a charge level beyond which it cannot generallyexceed1.

1This ‘asymptotic’ or ‘maximum’ charge level that pixels seem to reach is an interesting and potentiallyimportant topic on its own. For example, we have found that this level is nearly constant within any givenpixel column of the chip, and that this level can actually be exceeded under certain extreme circumstances.A full discussion of this topic is beyond the scope of this work.

4

The details of the pixel behavior observed in this figure (also see Fig. 3.9 of Howell (2006))are applicable to all charge-accumulating pixels (i.e. not just ‘super-hot’ pixels) and guide usin deciding how we should define ‘saturation’ for practical purposes, and how we quantify thesaturation onset. In the context of this work, we define the saturation level as the intensityof a charge-accumulating pixel at which bleeding begins (i.e. the intensity corresponding tothe first dashed line in Figure 2 for that particular pixel), which we differentiate from theintensity that the pixel asymptotically approaches at even higher charge levels/integrationtimes. Our goal then is to detect and measure the onset of this blooming and map it outover the WFC CCD area, as discussed in the following section. The details of the pixelbehavior gleaned from Figure 2, coupled with the full saturation map which we produce inthis work, are also crucial ingredients for accurate predictive modeling images of saturatedsources, which we plan to develop in future work (Section 4).

3 Saturation Map for the ACS/WFC

3.1 A Direct and Accurate Measurement of the Saturation Level

In order to measure the saturation level of a given pixel (or small group of pixels, assuminglocal variations are small) directly, we need to identify the charge level above which it starts tobleed charge into its neighboring pixels. For the case of individual super-hot pixels discussedin the previous section, we can rather easily fit/measure the saturation level in plots likeFigure 2. However, there is not a dense enough sampling of these objects over the detectorarea for an accurate and high resolution map (though they provide a coarse check on theproper saturation map that we construct in this section). Instead, we will make use of aslightly different analysis which still captures the behavior of interest. In particular, for alarge set of point sources (i.e. stars), we will plot the fraction of charge contained in theirpeak pixel as a function of the peak pixel intensity. Because of the excellent long-termstability of the ACS/WFC PSF, the reveals the saturation level as a sharp feature.

It may be instructive to elaborate on this approach: in the case of the super-hot pixels,assuming a background level of zero and other ideal conditions (e.g. no CTE losses, etc.),the total charge in a small region centered on the super-hot pixel is contributed entirelyby that source pixel. Thus, for all charge levels below the saturation level, i.e. before thepixel starts blooming, the fraction of the total charge (within that small region) containedin the source pixel would be unity. As the integration continues above the saturation level,the source pixel will bleed some charge into neighboring pixels (Figure 2), resulting in amonotonic decrease in the fraction of total charge contained in the source pixel – a featurewhich we can identify in the relevant plot (e.g. Figure 3).

The concept is similar for images of astronomical sources convolved with the PSF: Asthe central (or any) pixel begins to bleed, the fraction of the total charge associated withthat source contained in the particular pixel decreases. Crucially, since the PSF, whichessentially encodes the fraction of charge/light contained in a given pixel, is independentof total brightness of the (unsaturated) source, we can use a heterogeneous sample of PSF-convolved images of point sources spanning a range of intensities as a proxy for a singlesource imaged at a range of exposure times. (Identifying the onset of saturation using this

5

Figure 3: Fraction of charge contained in the peak pixel vs. the peak pixel intensity for stars imagedwithin a particular small (128 × 128 pixels) region of the WFC CCD area. Top: Fractional charge levelmeasured in background-subtracted images. Bottom: Fractional charge level measured in images with aconstant background level added in, as described in the text. Black points are measurements from over 500stars from ∼ 200 different images. Red and blue curves are a running median and polynomial fit of the data,respectively, and corresponding dashed lines running through their peaks. The abscissa of the two peaks arewithin 1% of each other, and we take their average value to be our estimate of the saturation level in thisparticular spatial bin. We automate this fitting and peak finding for all spatial bins over the detector areain order to map out the saturation level. This method provides the most direct and precise measurement ofthe saturation level out of any previously used methods.

method is in a sense the same as identifying the onset of PSF deviations.) An example ofthe relevant plot, for a certain small group of pixels on the WFC frame, is shown in Figure 3and is described further below.

In order to use these principles to accurately map out the saturation level over theWFC CCD area, we need a large sample of sources with dense sampling in brightness nearthe saturation level and dense sampling in position on the detector. For this, we leveragethe wealth of archival post-SM4 WFC data and perform the following procedure. First,we select an initial set of images consisting of over 200 individual F606W exposures of 47Tuc (full list given in the Appendix) spanning a range of exposure times between ∼ 40 sto ∼ 1400 s. These particular images were chosen due to excellent spatial and brightnesssampling. As mentioned in the Introduction, we use only full-frame images with gain=2. Weobtain the raw images and run acsccd on them in order to bias-subtract and convert the

6

Figure 4: Map of the saturation level over the WFC frame. The plotted area corresponds to the entireWFC CCD detector area with the two chips concatenated at the center. Left: The saturation level directlymeasured in spatial bins as described in the text. Right: Smooth polynomial fit to Left, resized to the fullresolution of WFC frames.

science frames to units of electrons. We then run Source Extractor (SE; Bertin & Arnouts,1996) on the science frames using mostly typical parameter values but slightly tweakedto optimize background subtraction and high fidelity source identification/measurement inthese crowded stellar fields. Along with the default output parameters, we tell SE to includein the output catalogs, for each source: the coordinates of the peak pixel, the peak pixelvalue, the elongation, the FWHM, and the stellarity index.

With the catalogs and images in hand, we then begin the saturation level analysis. Wefirst combine all the SE catalogs into a single catalog and clean it from likely cosmic raysand extended sources by cutting on the FWHM (< 5 pixels), stellarity index (> 0.8), andelongation (< 5). Since we know that the saturation level varies over the frame (Gilliland,2004), we bin the data by their image coordinates into bins 128×128 pixels large, a resolutionwhich represents a balance between ensuring a reasonable sample size in each bin and notsmoothing over too large of spatial scales. (This spatial binning assumes small local variationsin the saturation level – this assumption is later confirmed by examining diagnostic errormaps and using convergence testing.) For each bin, we can now generate a plot like Figure 3(top).

However, it is at this point that we must address a subtlety in the analysis. When usingproperly background-subtracted images for these measurements (as in Figure 3 (top)), thefractional charge for peak pixels leftwards of the saturation level will be scattered about aconstant value, with a somewhat gradual dropoff starting at the saturation level, as expectedby the nature of the PSF. The shape of this trend is not very conducive to automated iden-

7

tification of the dropoff since it does not represent a clear extrema. Fortunately, we canemploy a simple transformation to make this measurement easier. It is easy to convince one-self either analytically or using simulations that by adding a suitable constant ‘background’to all the pixels from which the total charge of each source is measured (i.e. when measuringthe quantity on the vertical axis of Figure 3 but NOT when measuring the quantity on thehorizontal axis), the shape of the trend will steepen leftwards of the dropoff, resulting inthe ordinate of the dropoff forming a clear peak without changing its abscissa. This makesit much easier to automate the identification of the saturation level by fitting the peak ofthe curve, as seen in the bottom panel of Figure 3. The value of the constant offset usedcan be tweaked by visually inspecting the relevant plot, ensuring that the desired behavioris achieved and that the relevant feature is preserved. We find that a large range of valuesworks fine for this, and we settle somewhat arbitrarily on a value of ∼ 8,000 electrons (∼ 1

10

of the typical saturation level) for this analysis.For each spatial bin, we then make a plot like the bottom panel of Figure 3. The total

charge of each source is measured within a 7× 7 pixel square aperture centered on the peakpixel. We fit the resulting scatter plot with a relatively low order polynomial (we find 8thorder to work well) and, separately, smooth it with a median filter. We then find the peakpixel intensity corresponding to the maximum of the polynomial and the median-filteredsignal, and we take their average intensity to be the saturation level for the given bin. Formost bins, the two values agree to within a few per cent, but for a handful of bins, there maybe large discrepancies, usually due to some large excursions of the polynomial fit near theboundaries of the fitting region or poor sampling of the data near the saturation level. Wecould inspect these bins by eye on a case-by-case basis and initialize the fit manually, butsince it only affects a small number of bins, it is inconsequential since we next fit a low-ordermodel to the whole map, which smoothes over these suspect bins. This smooth model isgenerated by splitting the resulting map (Figure 4 (left)) at the chip gap and fitting eachseparately with an 8th-order 2D polynomial (again, polynomial order chosen as an optimalvalue by eye). The two surfaces are then concatenated at the chip gap and the resultingimage is resized (using sklearn.transform.resize) to the proper pixel dimensions of WFCframes, namely 4096 × 4096 (Figure 4 (right)). The result of this final step is a map of thesaturation level which can readily be used to flag saturated pixels in WFC science framesthat have been processed with acsccd.

The mean saturation level over the whole map is ∼ 77,400 electrons, with peak-to-peakvariations of ± ∼ 5,000 electrons, or ± ∼ 6%, about the mean. For comparison, the valuecurrently used in the pipeline for flagging saturated pixels is 44,450 DN (corresponding to∼ 84,500 electrons given typical bias levels at GAIN=2), which falls entirely outside therange of our map. The non-uniform structure of the variations does not seem to be stronglyrelated to any other known structure inherent to the CCD, and arguably loosely resembles(qualitatively) the early result seen in Figure 4 of Gilliland (2004). We also find that thesaturation level has remained constant throughout the post-SM4 lifetime of the detector,precluding the need for any time-dependent terms/corrections.

8

3.2 Resulting Improved Saturated Pixel Flagging and Implemen-tation in the Pipeline

In order to test our new saturation map, we use it to flag saturated pixels in a varietyof science frames, including ones that were and were not used in the construction of themap. The flagging is done straightforwardly, just as it would be done in the calibrationpipeline, using a simple logical comparison – in all pixels where the value of the scienceframe is greater than the value of the saturation map, the saturation flag is turned on. Wecan then compare the results of this flagging with the current pipeline flagging by inspectingthe science frames and saturation masks side-by-side. Several examples of this comparisonare shown in Figure 5, from which it is clear that our new saturation map delivers superiorresults. From a sample of about 100 random science frames, we find that our new flaggingresults in ∼ 5% to ∼ 15% more pixels being flagged as saturated in any given frame, with amean of ∼ 9% more, compared to the current flagging.

There are likely several reasons for the inferior and sometimes unexpected flagging ofthe current pipeline: The current pipeline does not account for the spatial variation of thesaturation level over the detector area, instead using a single scalar threshold to check forsaturation over the entire frame. The value of this threshold is also simply too high andoften misses clearly saturated pixels belonging to extended saturation trails, or those formingthe (less so, but still saturated) boundaries of saturated sources, or otherwise. Moreover,this threshold is in units of DN, and currently the flagging is done before bias-subtractionand conversion to units of electrons using the gain. We caution that this approach may beimproper since the bias level is not constant between frames nor over the lifetime of thedetector (Desjardins & Khandrika, 2019), and since the gain varies by as much as ∼ 9%between the four CCD quadrants (Desjardins & Grogin, 2018).

Given the marked improvement of our new saturation flagging, the ACS instrument teamis now taking steps to implement it into the calacs pipeline, including changing the orderof operations in acsccd such that the saturation flagging happens after bias-subtraction andunit conversion. We hope to implement these updates into the pipeline in the near future.

Note: For gain settings lower than 2, full-well saturation cannot be automatically differ-entiated from A-to-D saturation using the methods of this work. Also, since gain settingsother than 2 are not supported by the ACS team since shortly after post-SM4, and thusvery few post-SM4 archival images using gain settings other than 2 exist, we plan to simplycontinue using a single scalar threshold for other gain settings unless indicated otherwise infuture documentation.

4 Extensions and Topics for Future Work

4.1 Full Blooming Model and HDR Darks

The current pipeline dark subtraction has a shortcoming related to the ‘super-hot’ pixels:When calibrating science frames, since the intensity of the long superdarks are scaled up ordown to match the exposure time of the science frame, frames with exposure times shorterthan ∼ 1000 s suffer from an oversubtraction of dark current in the pixels which form the

9

Figure 5: Visual comparison of several examples of our new saturated pixel flagging compared to thecurrent pipeline flagging. These particular example cutouts were chosen to showcase several different typesof situations where the current pipeline falls short. Left: Science image cutout. Middle left: Our newsaturation flagging. Middle right: Current pipeline saturation flagging. Right: Difference between new andcurrent saturation flagging, to guide the eye and highlight the current flagging shortcomings. All but thescience frames are simply boolean images: 1 (white) where the saturation flag is on and 0 (black) where itis not.

10

bleed trail and their immediate neighbors in the superdarks, but which are not yet bled intoin the science frame (see e.g. Miles (2018)). (The current pipeline also flags all the pixels inthe bleed trail as hot pixels in the DQ arrays, even though only the central pixel is truly ‘hot’in the sense that it is the source of the dark current.) The opposite issue occurs in scienceframes with exposure times larger than ∼ 1000 s, which suffer from an undersubtraction ofdark current in the pixels above and below the saturation trails. In order to properly subtractthe dark current associated with the saturation trails of hot pixels, either darks would needto be taken at multiple intermediate exposure times, or the saturation trails would need tobe modeled for a specific science frame exposure time. Since the former option is unviable,we turn to the latter option.

With our new saturation map, and a thorough understanding of the behavior of thesehot pixels and how they bleed their charge, it should be possible to accurately model thestructure and intensity of the dark current resulting from their saturation trails for any givenexposure time. Following this work, we plan to develop this tool, and use it to create so-called “High Dynamic Range” darks, which will feature fully modeled saturated hot pixelsand their bleed trails for any exposure times. A full blooming model as proposed would alsobe useful for simulating science images in which sources are expected to saturate. One mightexpect that the operating principle for this model should be as simple as spilling excesscharge beyond saturation into the nearest available pixel in the column, which it is to firstorder. However, there is a slight complication, discussed in the following subsection.

4.2 Possible Horizontal Blooming

The reader may have noticed a curious feature in the image cutouts from the long darksin Figure 1 – in addition to the familiar saturation blooming trail seen in the column ofthe source pixel, there also appears to be noticeable charge in the neighboring pixels onecolumn over in either direction. Some excess charge in the adjacent column on the sideopposite to readout is expected due to serial CTE losses, which are small (Anderson &Bedin, 2010), but currently uncorrected for. However, the charge observed in the columnon the other side, which is often much brighter than the CTE loss trail, is unexpectedand newly documented as of this work. This effect, which has evidently been present forthe entire lifetime of the detector, is also readily seen in science images of saturated stars(though it is harder to isolate due to nearby charge from any other sources), where the effectcan clearly appear on both sides of the saturation trail, and can span multiple columns. (Forreference, a similar effect is not observed in WFC3/UVIS images, though this comparisonmay be inconsequential because the two imagers use CCDs with different characteristicsand manufactured by different companies.) Our suspicion is that this excess charge is theresult of charge bleeding in the horizontal (serial) direction, though the conventional wisdomsuggests that this should not occur (Janesick, 2001). Another possibility is that this is relatedto charge diffusion (Krist, 2003), or perhaps that this charge could be bleeding during readout(i.e. due to some issue with the serial register) rather than during integration. Regardless ofit’s nature, we have spent significant effort studying the phenomenon in WFC images, whichhas resulted in a significant, but as of yet incomplete, characterization of it. We plan todiscuss this topic in detail in future work, but suffice it to say for the moment that the waysin which this effect manifests and behaves are considerably unlike familiar vertical blooming,

11

and defies the straightforward modeling that successfully quantifies WFC vertical blooming.

4.3 Pre-SM4 Analysis

The results presented in this work apply only to post-SM4 WFC data. There may bereason to suspect that the saturation levels were different pre-SM4 due to the differentreadout electronics and settings used. In the future we plan to carry out this analysis forpre-SM4 data.

5 Summary

The current ACS/WFC calibration pipeline’s saturated pixel flagging routine producessomewhat unsatisfactory and unexpected results, failing to identify many clearly saturatedpixels. We have developed and deployed a new method for measuring the position-dependentsaturation level of the ACS/WFC CCD (and in principle, any CCD). This method, whichoffers both a more direct and more precise measurement compared to the previously usedmethods, operates by identifying pixels which are just beginning to bleed charge into theirneighboring pixels. We leverage a large sample of post-SM4 archival image data in order toachieve a high precision measurement of this effect, and map the saturation level across theCCDs. We find that our resulting map of the saturation level provides results significantlysuperior to the current pipeline flagging, and we plan to implement it into the calibrationpipeline in the near future. Several related issues yet remain unsolved, and we plan to explorethem in future work.

Acknowledgements

We thank Nathan Miles, Tyler Desjardins, Melanie Olaes, and the whole ACS team forilluminating discussions and support of this work.

This research made use of: NumPy (Van Der Walt et al., 2011), SciPy (Virtanen et al.,2019), matplotlib (Hunter, 2007), Astropy (Price-Whelan et al., 2018), and Scikit-learn(McKinney, 2010).

References

Anderson, J., & Bedin, L. 2010, An Empirical Pixel-Based Correction for Imperfect CTE. I.HST’s Advanced Camera for Surveys, Tech. rep.

Bertin, E., & Arnouts, S. 1996, Astronomy and Astrophysics Supplement, 117, 393

Desjardins, T. D., & Grogin, N. A. 2018, Remeasuring the ACS/WFC Absolute Gains, Tech.rep.

Desjardins, T. D., & Khandrika, H. G. 2019, Post-SM4 ACS/WFC Bias II: Temporal Struc-ture in the Prescan Bias Level, Tech. rep.

12

Gilliland, R. L. 2004, ACS CCD Gains, Full Well Depths, and Linearity up to and BeyondSaturation, Tech. rep.

Golimowski, D., Cheng, E., Loose, M., et al. 2011, ACS after Servicing Mission 4: The WFCOptimization Campaign, Tech. Rep. ACS ISR 2011-04, STScI

Howell, S. B. 2006, Handbook of CCD Astronomy, Vol. 5

Hunter, J. D. 2007, Computing In Science & Engineering, 9, 90

Janesick, J. R. 2001, Scientific charge-coupled devices

Krist, J. 2003, ACS WFC &amp; HRC fielddependent PSF variations due to optical andcharge diffusion effects, Tech. rep.

Lucas et al. 2018, ACS Data Handbook, Version 9.0

McKinney, W. 2010, in Proceedings of the 9th Python in Science Conference, ed. S. van derWalt & J. Millman, 51 – 56

Miles, N. 2018, Updates to Post-Flash Calibration for the Advanced Camera for SurveysWide Field Channel, Tech. rep.

Ogaz, S., Anderson, J., & Golimowski, D. 2015, Post-Flash Calibration Darks for the Ad-vanced Camera for Surveys Wide Field Channel (ACS/WFC), Tech. rep.

Price-Whelan, A. M., Sipocz, B. M., Gunther, H. M., et al. 2018, Astronomical Journal, 156,123

Riess, A., Mutchler, M., & van Orsow, D. 2002, A First Look at Hot Pixels on ACS, Tech.rep.

Ryon et al. 2019, Advanced Camera for Surveys Instrument Handbook for Cycle 27 v. 18.0

Van Der Walt, S., Colbert, S. C., & Varoquaux, G. 2011, Computing in Science & Engineer-ing, 13, 22

Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2019, arXiv e-prints, arXiv:1907.10121

Appendix

List of ACS/WFC datasets used in this work:ja9702spq, ja9702srq, ja9702sxq, ja9702szq, ja9702t5q, ja9702t7q, ja9bw2a4q, ja9bw2a5q,ja9bw2a7q, ja9bw2a9q, ja9bw2abq, ja9bw2adq, ja9bw2ykq, ja9bw2ylq, ja9bw2zdq, ja9bw2zfq,ja9bw2zhq, ja9bw2zjq, ja9bw2zqq, ja9bw2zsq, ja9bw2zuq, ja9bw2zwq, jb6v01diq, jb6v01dkq,jb6v01e0q, jb6v01e5q, jb6v01ekq, jb6v02x7q, jb6v02xbq, jb6v02xhq, jb6v02xmq, jb6v02zuq,jb6v03hfq, jb6v03hlq, jb6v03i2q, jb6v03i7q, jb6v03imq, jb6v04a4q, jb6v04a9q, jb6v04aoq,jb6v04ztq, jb6v04zxq, jb6v06bzq, jb6v06c3q, jb6v06c9q, jb6v06ceq, jb6v06csq, jb6v07wmq,

13

jb6v07wpq, jb6v07x5q, jb6v07xaq, jb6v07xpq, jb6v08xjq, jb6v08xoq, jb6v08y3q, jb6v09shq,jb6v09sjq, jb6v09szq, jb6v09t4q, jb6v09tjq, jb6v10njq, jb6v10nnq, jb6v10ntq, jb6v10nyq,jb6v10odq, jb6v11kyq, jb6v11l0q, jb6v11lgq, jb6v11llq, jb6v11m0q, jb6v12mrq, jb6v12mvq,jb6v12n1q, jb6v12n6q, jb6v12nlq, jb6v13y2q, jb6v13y4q, jb6v13ykq, jb6v13ypq, jb6v13z4q,jb6v14d9q, jb6v14dcq, jb6v14dgq, jb6v14dlq, jb6v14e0q, jb6v15amq, jb6v15bkq, jb6v15c9q,jb6v15ceq, jb6v15d4q, jb6v16hvq, jb6v16hyq, jb6v16i4q, jb6v16idq, jb6v16itq, jb6v17n0q,jb6v17ngq, jb6v17nlq, jb6v17o0q, jb6v18siq, jb6v18slq, jb6v18suq, jb6v18szq, jb6v18teq,jb6v19byq, jb6v19coq, jb6v19cyq, jb6v19ddq, jb6v20drq, jb6v20e1q, jb6v20egq, jb6v21f5q,jb6v21f7q, jb6v21g4q, jb6v21gaq, jb6v21hhq, jb6v22lhq, jb6v22lkq, jb6v22loq, jb6v22ltq,jb6v22m8q, jb6v23puq, jb6v23qvq, jb6v23r0q, jb6v23req, jb6v24udq, jb6v24ugq, jb6v24ukq,jb6v24upq, jb6v24v4q, jb6v24veq, jb6v25adq, jb6v25afq, jb6v25azq, jb6v25baq, jb6v25csq,jbb501f9q, jbb501fbq, jbb501glq, jbb501heq, jbb501hgq, jbb502i8q, jbb502iaq, jbb502igq,jbb502ijq, jbb502ipq, jbb502irq, jbbf01i3q, jbbfw2ikq, jbbfw5eqq, jbms01kxq, jbms02fgq,jbms03olq, jbn501pqq, jbn501psq, jbn501q3q, jbn501q5q, jbn501qcq, jbn501qeq, jbn502qhq,jbn502qjq, jbn502qpq, jbn502qrq, jbn502qxq, jbn502qzq, jbva01mwq, jbva02krq, jbva03jcq,jbw801xjq, jbw801xlq, jbw801xtq, jbw801xvq, jbw801y1q, jbw801y3q, jbw802ybq, jbw802ydq,jbw802yjq, jbw802ylq, jbw802yrq, jbw802ytq, jc5001s9q, jc5002olq, jc5003fpq, jc6101h6q,jc6101h8q, jc6101hhq, jc6101hjq, jc6101hzq, jc6101i1q, jc6102ixq, jc6102izq, jc6102jaq, jc6102jyq,jc6102k4q, jc6102k6q, jce501e9q, jce502hoq, jce503bkq, jcfm01plq, jcfm01pnq, jcfm01ptq,jcfm01pvq, jcfm01q1q, jcfm01q3q, jcfm02qaq, jcfm02qcq, jcfm02qiq, jcfm02qkq, jcfm02qmq,jcfm02qoq, jcr513hcq

14